4 * (C) 1997 Linus Torvalds
9 #include <linux/dcache.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/writeback.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/wait.h>
16 #include <linux/rwsem.h>
17 #include <linux/hash.h>
18 #include <linux/swap.h>
19 #include <linux/security.h>
20 #include <linux/pagemap.h>
21 #include <linux/cdev.h>
22 #include <linux/bootmem.h>
23 #include <linux/inotify.h>
24 #include <linux/fsnotify.h>
25 #include <linux/mount.h>
26 #include <linux/async.h>
27 #include <linux/posix_acl.h>
30 * This is needed for the following functions:
32 * - invalidate_inode_buffers
35 * FIXME: remove all knowledge of the buffer layer from this file
37 #include <linux/buffer_head.h>
40 * New inode.c implementation.
42 * This implementation has the basic premise of trying
43 * to be extremely low-overhead and SMP-safe, yet be
44 * simple enough to be "obviously correct".
49 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
51 /* #define INODE_PARANOIA 1 */
52 /* #define INODE_DEBUG 1 */
55 * Inode lookup is no longer as critical as it used to be:
56 * most of the lookups are going to be through the dcache.
58 #define I_HASHBITS i_hash_shift
59 #define I_HASHMASK i_hash_mask
61 static unsigned int i_hash_mask __read_mostly
;
62 static unsigned int i_hash_shift __read_mostly
;
65 * Each inode can be on two separate lists. One is
66 * the hash list of the inode, used for lookups. The
67 * other linked list is the "type" list:
68 * "in_use" - valid inode, i_count > 0, i_nlink > 0
69 * "dirty" - as "in_use" but also dirty
70 * "unused" - valid inode, i_count = 0
72 * A "dirty" list is maintained for each super block,
73 * allowing for low-overhead inode sync() operations.
76 LIST_HEAD(inode_in_use
);
77 LIST_HEAD(inode_unused
);
78 static struct hlist_head
*inode_hashtable __read_mostly
;
81 * A simple spinlock to protect the list manipulations.
83 * NOTE! You also have to own the lock if you change
84 * the i_state of an inode while it is in use..
86 DEFINE_SPINLOCK(inode_lock
);
89 * iprune_sem provides exclusion between the kswapd or try_to_free_pages
90 * icache shrinking path, and the umount path. Without this exclusion,
91 * by the time prune_icache calls iput for the inode whose pages it has
92 * been invalidating, or by the time it calls clear_inode & destroy_inode
93 * from its final dispose_list, the struct super_block they refer to
94 * (for inode->i_sb->s_op) may already have been freed and reused.
96 * We make this an rwsem because the fastpath is icache shrinking. In
97 * some cases a filesystem may be doing a significant amount of work in
98 * its inode reclaim code, so this should improve parallelism.
100 static DECLARE_RWSEM(iprune_sem
);
103 * Statistics gathering..
105 struct inodes_stat_t inodes_stat
;
107 static struct kmem_cache
*inode_cachep __read_mostly
;
109 static void wake_up_inode(struct inode
*inode
)
112 * Prevent speculative execution through spin_unlock(&inode_lock);
115 wake_up_bit(&inode
->i_state
, __I_NEW
);
119 * inode_init_always - perform inode structure intialisation
120 * @sb: superblock inode belongs to
121 * @inode: inode to initialise
123 * These are initializations that need to be done on every inode
124 * allocation as the fields are not initialised by slab allocation.
126 int inode_init_always(struct super_block
*sb
, struct inode
*inode
)
128 static const struct address_space_operations empty_aops
;
129 static const struct inode_operations empty_iops
;
130 static const struct file_operations empty_fops
;
131 struct address_space
*const mapping
= &inode
->i_data
;
134 inode
->i_blkbits
= sb
->s_blocksize_bits
;
136 atomic_set(&inode
->i_count
, 1);
137 inode
->i_op
= &empty_iops
;
138 inode
->i_fop
= &empty_fops
;
142 atomic_set(&inode
->i_writecount
, 0);
146 inode
->i_generation
= 0;
148 memset(&inode
->i_dquot
, 0, sizeof(inode
->i_dquot
));
150 inode
->i_pipe
= NULL
;
151 inode
->i_bdev
= NULL
;
152 inode
->i_cdev
= NULL
;
154 inode
->dirtied_when
= 0;
156 if (security_inode_alloc(inode
))
158 spin_lock_init(&inode
->i_lock
);
159 lockdep_set_class(&inode
->i_lock
, &sb
->s_type
->i_lock_key
);
161 mutex_init(&inode
->i_mutex
);
162 lockdep_set_class(&inode
->i_mutex
, &sb
->s_type
->i_mutex_key
);
164 init_rwsem(&inode
->i_alloc_sem
);
165 lockdep_set_class(&inode
->i_alloc_sem
, &sb
->s_type
->i_alloc_sem_key
);
167 mapping
->a_ops
= &empty_aops
;
168 mapping
->host
= inode
;
170 mapping_set_gfp_mask(mapping
, GFP_HIGHUSER_MOVABLE
);
171 mapping
->assoc_mapping
= NULL
;
172 mapping
->backing_dev_info
= &default_backing_dev_info
;
173 mapping
->writeback_index
= 0;
176 * If the block_device provides a backing_dev_info for client
177 * inodes then use that. Otherwise the inode share the bdev's
181 struct backing_dev_info
*bdi
;
183 bdi
= sb
->s_bdev
->bd_inode
->i_mapping
->backing_dev_info
;
184 mapping
->backing_dev_info
= bdi
;
186 inode
->i_private
= NULL
;
187 inode
->i_mapping
= mapping
;
188 #ifdef CONFIG_FS_POSIX_ACL
189 inode
->i_acl
= inode
->i_default_acl
= ACL_NOT_CACHED
;
192 #ifdef CONFIG_FSNOTIFY
193 inode
->i_fsnotify_mask
= 0;
200 EXPORT_SYMBOL(inode_init_always
);
202 static struct inode
*alloc_inode(struct super_block
*sb
)
206 if (sb
->s_op
->alloc_inode
)
207 inode
= sb
->s_op
->alloc_inode(sb
);
209 inode
= kmem_cache_alloc(inode_cachep
, GFP_KERNEL
);
214 if (unlikely(inode_init_always(sb
, inode
))) {
215 if (inode
->i_sb
->s_op
->destroy_inode
)
216 inode
->i_sb
->s_op
->destroy_inode(inode
);
218 kmem_cache_free(inode_cachep
, inode
);
225 void __destroy_inode(struct inode
*inode
)
227 BUG_ON(inode_has_buffers(inode
));
228 security_inode_free(inode
);
229 fsnotify_inode_delete(inode
);
230 #ifdef CONFIG_FS_POSIX_ACL
231 if (inode
->i_acl
&& inode
->i_acl
!= ACL_NOT_CACHED
)
232 posix_acl_release(inode
->i_acl
);
233 if (inode
->i_default_acl
&& inode
->i_default_acl
!= ACL_NOT_CACHED
)
234 posix_acl_release(inode
->i_default_acl
);
237 EXPORT_SYMBOL(__destroy_inode
);
239 void destroy_inode(struct inode
*inode
)
241 __destroy_inode(inode
);
242 if (inode
->i_sb
->s_op
->destroy_inode
)
243 inode
->i_sb
->s_op
->destroy_inode(inode
);
245 kmem_cache_free(inode_cachep
, (inode
));
248 void address_space_init_once(struct address_space
*mapping
)
250 memset(mapping
, 0, sizeof(*mapping
));
251 INIT_RADIX_TREE(&mapping
->page_tree
, GFP_ATOMIC
);
252 spin_lock_init(&mapping
->tree_lock
);
253 spin_lock_init(&mapping
->i_mmap_lock
);
254 INIT_LIST_HEAD(&mapping
->private_list
);
255 spin_lock_init(&mapping
->private_lock
);
256 INIT_RAW_PRIO_TREE_ROOT(&mapping
->i_mmap
);
257 INIT_LIST_HEAD(&mapping
->i_mmap_nonlinear
);
258 mutex_init(&mapping
->unmap_mutex
);
260 EXPORT_SYMBOL(address_space_init_once
);
263 * These are initializations that only need to be done
264 * once, because the fields are idempotent across use
265 * of the inode, so let the slab aware of that.
267 void inode_init_once(struct inode
*inode
)
269 memset(inode
, 0, sizeof(*inode
));
270 INIT_HLIST_NODE(&inode
->i_hash
);
271 INIT_LIST_HEAD(&inode
->i_dentry
);
272 INIT_LIST_HEAD(&inode
->i_devices
);
273 address_space_init_once(&inode
->i_data
);
274 i_size_ordered_init(inode
);
275 #ifdef CONFIG_INOTIFY
276 INIT_LIST_HEAD(&inode
->inotify_watches
);
277 mutex_init(&inode
->inotify_mutex
);
279 #ifdef CONFIG_FSNOTIFY
280 INIT_HLIST_HEAD(&inode
->i_fsnotify_mark_entries
);
283 EXPORT_SYMBOL(inode_init_once
);
285 static void init_once(void *foo
)
287 struct inode
*inode
= (struct inode
*) foo
;
289 inode_init_once(inode
);
293 * inode_lock must be held
295 void __iget(struct inode
*inode
)
297 if (atomic_inc_return(&inode
->i_count
) != 1)
300 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
301 list_move(&inode
->i_list
, &inode_in_use
);
302 inodes_stat
.nr_unused
--;
306 * clear_inode - clear an inode
307 * @inode: inode to clear
309 * This is called by the filesystem to tell us
310 * that the inode is no longer useful. We just
311 * terminate it with extreme prejudice.
313 void clear_inode(struct inode
*inode
)
316 invalidate_inode_buffers(inode
);
318 BUG_ON(inode
->i_data
.nrpages
);
319 BUG_ON(!(inode
->i_state
& I_FREEING
));
320 BUG_ON(inode
->i_state
& I_CLEAR
);
321 inode_sync_wait(inode
);
322 if (inode
->i_sb
->s_op
->clear_inode
)
323 inode
->i_sb
->s_op
->clear_inode(inode
);
324 if (S_ISBLK(inode
->i_mode
) && inode
->i_bdev
)
326 if (S_ISCHR(inode
->i_mode
) && inode
->i_cdev
)
328 inode
->i_state
= I_CLEAR
;
330 EXPORT_SYMBOL(clear_inode
);
333 * dispose_list - dispose of the contents of a local list
334 * @head: the head of the list to free
336 * Dispose-list gets a local list with local inodes in it, so it doesn't
337 * need to worry about list corruption and SMP locks.
339 static void dispose_list(struct list_head
*head
)
343 while (!list_empty(head
)) {
346 inode
= list_first_entry(head
, struct inode
, i_list
);
347 list_del(&inode
->i_list
);
349 if (inode
->i_data
.nrpages
)
350 truncate_inode_pages(&inode
->i_data
, 0);
353 spin_lock(&inode_lock
);
354 hlist_del_init(&inode
->i_hash
);
355 list_del_init(&inode
->i_sb_list
);
356 spin_unlock(&inode_lock
);
358 wake_up_inode(inode
);
359 destroy_inode(inode
);
362 spin_lock(&inode_lock
);
363 inodes_stat
.nr_inodes
-= nr_disposed
;
364 spin_unlock(&inode_lock
);
368 * Invalidate all inodes for a device.
370 static int invalidate_list(struct list_head
*head
, struct list_head
*dispose
)
372 struct list_head
*next
;
373 int busy
= 0, count
= 0;
377 struct list_head
*tmp
= next
;
381 * We can reschedule here without worrying about the list's
382 * consistency because the per-sb list of inodes must not
383 * change during umount anymore, and because iprune_sem keeps
384 * shrink_icache_memory() away.
386 cond_resched_lock(&inode_lock
);
391 inode
= list_entry(tmp
, struct inode
, i_sb_list
);
392 if (inode
->i_state
& I_NEW
)
394 invalidate_inode_buffers(inode
);
395 if (!atomic_read(&inode
->i_count
)) {
396 list_move(&inode
->i_list
, dispose
);
397 WARN_ON(inode
->i_state
& I_NEW
);
398 inode
->i_state
|= I_FREEING
;
404 /* only unused inodes may be cached with i_count zero */
405 inodes_stat
.nr_unused
-= count
;
410 * invalidate_inodes - discard the inodes on a device
413 * Discard all of the inodes for a given superblock. If the discard
414 * fails because there are busy inodes then a non zero value is returned.
415 * If the discard is successful all the inodes have been discarded.
417 int invalidate_inodes(struct super_block
*sb
)
420 LIST_HEAD(throw_away
);
422 down_write(&iprune_sem
);
423 spin_lock(&inode_lock
);
424 inotify_unmount_inodes(&sb
->s_inodes
);
425 fsnotify_unmount_inodes(&sb
->s_inodes
);
426 busy
= invalidate_list(&sb
->s_inodes
, &throw_away
);
427 spin_unlock(&inode_lock
);
429 dispose_list(&throw_away
);
430 up_write(&iprune_sem
);
434 EXPORT_SYMBOL(invalidate_inodes
);
436 static int can_unuse(struct inode
*inode
)
440 if (inode_has_buffers(inode
))
442 if (atomic_read(&inode
->i_count
))
444 if (inode
->i_data
.nrpages
)
450 * Scan `goal' inodes on the unused list for freeable ones. They are moved to
451 * a temporary list and then are freed outside inode_lock by dispose_list().
453 * Any inodes which are pinned purely because of attached pagecache have their
454 * pagecache removed. We expect the final iput() on that inode to add it to
455 * the front of the inode_unused list. So look for it there and if the
456 * inode is still freeable, proceed. The right inode is found 99.9% of the
457 * time in testing on a 4-way.
459 * If the inode has metadata buffers attached to mapping->private_list then
460 * try to remove them.
462 static void prune_icache(int nr_to_scan
)
467 unsigned long reap
= 0;
469 down_read(&iprune_sem
);
470 spin_lock(&inode_lock
);
471 for (nr_scanned
= 0; nr_scanned
< nr_to_scan
; nr_scanned
++) {
474 if (list_empty(&inode_unused
))
477 inode
= list_entry(inode_unused
.prev
, struct inode
, i_list
);
479 if (inode
->i_state
|| atomic_read(&inode
->i_count
)) {
480 list_move(&inode
->i_list
, &inode_unused
);
483 if (inode_has_buffers(inode
) || inode
->i_data
.nrpages
) {
485 spin_unlock(&inode_lock
);
486 if (remove_inode_buffers(inode
))
487 reap
+= invalidate_mapping_pages(&inode
->i_data
,
490 spin_lock(&inode_lock
);
492 if (inode
!= list_entry(inode_unused
.next
,
493 struct inode
, i_list
))
494 continue; /* wrong inode or list_empty */
495 if (!can_unuse(inode
))
498 list_move(&inode
->i_list
, &freeable
);
499 WARN_ON(inode
->i_state
& I_NEW
);
500 inode
->i_state
|= I_FREEING
;
503 inodes_stat
.nr_unused
-= nr_pruned
;
504 if (current_is_kswapd())
505 __count_vm_events(KSWAPD_INODESTEAL
, reap
);
507 __count_vm_events(PGINODESTEAL
, reap
);
508 spin_unlock(&inode_lock
);
510 dispose_list(&freeable
);
511 up_read(&iprune_sem
);
515 * shrink_icache_memory() will attempt to reclaim some unused inodes. Here,
516 * "unused" means that no dentries are referring to the inodes: the files are
517 * not open and the dcache references to those inodes have already been
520 * This function is passed the number of inodes to scan, and it returns the
521 * total number of remaining possibly-reclaimable inodes.
523 static int shrink_icache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
527 * Nasty deadlock avoidance. We may hold various FS locks,
528 * and we don't want to recurse into the FS that called us
529 * in clear_inode() and friends..
531 if (!(gfp_mask
& __GFP_FS
))
535 return (inodes_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
538 static struct shrinker icache_shrinker
= {
539 .shrink
= shrink_icache_memory
,
540 .seeks
= DEFAULT_SEEKS
,
543 static void __wait_on_freeing_inode(struct inode
*inode
);
545 * Called with the inode lock held.
546 * NOTE: we are not increasing the inode-refcount, you must call __iget()
547 * by hand after calling find_inode now! This simplifies iunique and won't
548 * add any additional branch in the common code.
550 static struct inode
*find_inode(struct super_block
*sb
,
551 struct hlist_head
*head
,
552 int (*test
)(struct inode
*, void *),
555 struct hlist_node
*node
;
556 struct inode
*inode
= NULL
;
559 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
560 if (inode
->i_sb
!= sb
)
562 if (!test(inode
, data
))
564 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
565 __wait_on_freeing_inode(inode
);
570 return node
? inode
: NULL
;
574 * find_inode_fast is the fast path version of find_inode, see the comment at
575 * iget_locked for details.
577 static struct inode
*find_inode_fast(struct super_block
*sb
,
578 struct hlist_head
*head
, unsigned long ino
)
580 struct hlist_node
*node
;
581 struct inode
*inode
= NULL
;
584 hlist_for_each_entry(inode
, node
, head
, i_hash
) {
585 if (inode
->i_ino
!= ino
)
587 if (inode
->i_sb
!= sb
)
589 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)) {
590 __wait_on_freeing_inode(inode
);
595 return node
? inode
: NULL
;
598 static unsigned long hash(struct super_block
*sb
, unsigned long hashval
)
602 tmp
= (hashval
* (unsigned long)sb
) ^ (GOLDEN_RATIO_PRIME
+ hashval
) /
604 tmp
= tmp
^ ((tmp
^ GOLDEN_RATIO_PRIME
) >> I_HASHBITS
);
605 return tmp
& I_HASHMASK
;
609 __inode_add_to_lists(struct super_block
*sb
, struct hlist_head
*head
,
612 inodes_stat
.nr_inodes
++;
613 list_add(&inode
->i_list
, &inode_in_use
);
614 list_add(&inode
->i_sb_list
, &sb
->s_inodes
);
616 hlist_add_head(&inode
->i_hash
, head
);
620 * inode_add_to_lists - add a new inode to relevant lists
621 * @sb: superblock inode belongs to
622 * @inode: inode to mark in use
624 * When an inode is allocated it needs to be accounted for, added to the in use
625 * list, the owning superblock and the inode hash. This needs to be done under
626 * the inode_lock, so export a function to do this rather than the inode lock
627 * itself. We calculate the hash list to add to here so it is all internal
628 * which requires the caller to have already set up the inode number in the
631 void inode_add_to_lists(struct super_block
*sb
, struct inode
*inode
)
633 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, inode
->i_ino
);
635 spin_lock(&inode_lock
);
636 __inode_add_to_lists(sb
, head
, inode
);
637 spin_unlock(&inode_lock
);
639 EXPORT_SYMBOL_GPL(inode_add_to_lists
);
642 * new_inode - obtain an inode
645 * Allocates a new inode for given superblock. The default gfp_mask
646 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
647 * If HIGHMEM pages are unsuitable or it is known that pages allocated
648 * for the page cache are not reclaimable or migratable,
649 * mapping_set_gfp_mask() must be called with suitable flags on the
650 * newly created inode's mapping
653 struct inode
*new_inode(struct super_block
*sb
)
656 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
657 * error if st_ino won't fit in target struct field. Use 32bit counter
658 * here to attempt to avoid that.
660 static unsigned int last_ino
;
663 spin_lock_prefetch(&inode_lock
);
665 inode
= alloc_inode(sb
);
667 spin_lock(&inode_lock
);
668 __inode_add_to_lists(sb
, NULL
, inode
);
669 inode
->i_ino
= ++last_ino
;
671 spin_unlock(&inode_lock
);
675 EXPORT_SYMBOL(new_inode
);
677 void unlock_new_inode(struct inode
*inode
)
679 #ifdef CONFIG_DEBUG_LOCK_ALLOC
680 if (inode
->i_mode
& S_IFDIR
) {
681 struct file_system_type
*type
= inode
->i_sb
->s_type
;
683 /* Set new key only if filesystem hasn't already changed it */
684 if (!lockdep_match_class(&inode
->i_mutex
,
685 &type
->i_mutex_key
)) {
687 * ensure nobody is actually holding i_mutex
689 mutex_destroy(&inode
->i_mutex
);
690 mutex_init(&inode
->i_mutex
);
691 lockdep_set_class(&inode
->i_mutex
,
692 &type
->i_mutex_dir_key
);
697 * This is special! We do not need the spinlock when clearing I_NEW,
698 * because we're guaranteed that nobody else tries to do anything about
699 * the state of the inode when it is locked, as we just created it (so
700 * there can be no old holders that haven't tested I_NEW).
701 * However we must emit the memory barrier so that other CPUs reliably
702 * see the clearing of I_NEW after the other inode initialisation has
706 WARN_ON(!(inode
->i_state
& I_NEW
));
707 inode
->i_state
&= ~I_NEW
;
708 wake_up_inode(inode
);
710 EXPORT_SYMBOL(unlock_new_inode
);
713 * This is called without the inode lock held.. Be careful.
715 * We no longer cache the sb_flags in i_flags - see fs.h
716 * -- rmk@arm.uk.linux.org
718 static struct inode
*get_new_inode(struct super_block
*sb
,
719 struct hlist_head
*head
,
720 int (*test
)(struct inode
*, void *),
721 int (*set
)(struct inode
*, void *),
726 inode
= alloc_inode(sb
);
730 spin_lock(&inode_lock
);
731 /* We released the lock, so.. */
732 old
= find_inode(sb
, head
, test
, data
);
734 if (set(inode
, data
))
737 __inode_add_to_lists(sb
, head
, inode
);
738 inode
->i_state
= I_NEW
;
739 spin_unlock(&inode_lock
);
741 /* Return the locked inode with I_NEW set, the
742 * caller is responsible for filling in the contents
748 * Uhhuh, somebody else created the same inode under
749 * us. Use the old inode instead of the one we just
753 spin_unlock(&inode_lock
);
754 destroy_inode(inode
);
756 wait_on_inode(inode
);
761 spin_unlock(&inode_lock
);
762 destroy_inode(inode
);
767 * get_new_inode_fast is the fast path version of get_new_inode, see the
768 * comment at iget_locked for details.
770 static struct inode
*get_new_inode_fast(struct super_block
*sb
,
771 struct hlist_head
*head
, unsigned long ino
)
775 inode
= alloc_inode(sb
);
779 spin_lock(&inode_lock
);
780 /* We released the lock, so.. */
781 old
= find_inode_fast(sb
, head
, ino
);
784 __inode_add_to_lists(sb
, head
, inode
);
785 inode
->i_state
= I_NEW
;
786 spin_unlock(&inode_lock
);
788 /* Return the locked inode with I_NEW set, the
789 * caller is responsible for filling in the contents
795 * Uhhuh, somebody else created the same inode under
796 * us. Use the old inode instead of the one we just
800 spin_unlock(&inode_lock
);
801 destroy_inode(inode
);
803 wait_on_inode(inode
);
809 * iunique - get a unique inode number
811 * @max_reserved: highest reserved inode number
813 * Obtain an inode number that is unique on the system for a given
814 * superblock. This is used by file systems that have no natural
815 * permanent inode numbering system. An inode number is returned that
816 * is higher than the reserved limit but unique.
819 * With a large number of inodes live on the file system this function
820 * currently becomes quite slow.
822 ino_t
iunique(struct super_block
*sb
, ino_t max_reserved
)
825 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
826 * error if st_ino won't fit in target struct field. Use 32bit counter
827 * here to attempt to avoid that.
829 static unsigned int counter
;
831 struct hlist_head
*head
;
834 spin_lock(&inode_lock
);
836 if (counter
<= max_reserved
)
837 counter
= max_reserved
+ 1;
839 head
= inode_hashtable
+ hash(sb
, res
);
840 inode
= find_inode_fast(sb
, head
, res
);
841 } while (inode
!= NULL
);
842 spin_unlock(&inode_lock
);
846 EXPORT_SYMBOL(iunique
);
848 struct inode
*igrab(struct inode
*inode
)
850 spin_lock(&inode_lock
);
851 if (!(inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
)))
855 * Handle the case where s_op->clear_inode is not been
856 * called yet, and somebody is calling igrab
857 * while the inode is getting freed.
860 spin_unlock(&inode_lock
);
863 EXPORT_SYMBOL(igrab
);
866 * ifind - internal function, you want ilookup5() or iget5().
867 * @sb: super block of file system to search
868 * @head: the head of the list to search
869 * @test: callback used for comparisons between inodes
870 * @data: opaque data pointer to pass to @test
871 * @wait: if true wait for the inode to be unlocked, if false do not
873 * ifind() searches for the inode specified by @data in the inode
874 * cache. This is a generalized version of ifind_fast() for file systems where
875 * the inode number is not sufficient for unique identification of an inode.
877 * If the inode is in the cache, the inode is returned with an incremented
880 * Otherwise NULL is returned.
882 * Note, @test is called with the inode_lock held, so can't sleep.
884 static struct inode
*ifind(struct super_block
*sb
,
885 struct hlist_head
*head
, int (*test
)(struct inode
*, void *),
886 void *data
, const int wait
)
890 spin_lock(&inode_lock
);
891 inode
= find_inode(sb
, head
, test
, data
);
894 spin_unlock(&inode_lock
);
896 wait_on_inode(inode
);
899 spin_unlock(&inode_lock
);
904 * ifind_fast - internal function, you want ilookup() or iget().
905 * @sb: super block of file system to search
906 * @head: head of the list to search
907 * @ino: inode number to search for
909 * ifind_fast() searches for the inode @ino in the inode cache. This is for
910 * file systems where the inode number is sufficient for unique identification
913 * If the inode is in the cache, the inode is returned with an incremented
916 * Otherwise NULL is returned.
918 static struct inode
*ifind_fast(struct super_block
*sb
,
919 struct hlist_head
*head
, unsigned long ino
)
923 spin_lock(&inode_lock
);
924 inode
= find_inode_fast(sb
, head
, ino
);
927 spin_unlock(&inode_lock
);
928 wait_on_inode(inode
);
931 spin_unlock(&inode_lock
);
936 * ilookup5_nowait - search for an inode in the inode cache
937 * @sb: super block of file system to search
938 * @hashval: hash value (usually inode number) to search for
939 * @test: callback used for comparisons between inodes
940 * @data: opaque data pointer to pass to @test
942 * ilookup5() uses ifind() to search for the inode specified by @hashval and
943 * @data in the inode cache. This is a generalized version of ilookup() for
944 * file systems where the inode number is not sufficient for unique
945 * identification of an inode.
947 * If the inode is in the cache, the inode is returned with an incremented
948 * reference count. Note, the inode lock is not waited upon so you have to be
949 * very careful what you do with the returned inode. You probably should be
950 * using ilookup5() instead.
952 * Otherwise NULL is returned.
954 * Note, @test is called with the inode_lock held, so can't sleep.
956 struct inode
*ilookup5_nowait(struct super_block
*sb
, unsigned long hashval
,
957 int (*test
)(struct inode
*, void *), void *data
)
959 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
961 return ifind(sb
, head
, test
, data
, 0);
963 EXPORT_SYMBOL(ilookup5_nowait
);
966 * ilookup5 - search for an inode in the inode cache
967 * @sb: super block of file system to search
968 * @hashval: hash value (usually inode number) to search for
969 * @test: callback used for comparisons between inodes
970 * @data: opaque data pointer to pass to @test
972 * ilookup5() uses ifind() to search for the inode specified by @hashval and
973 * @data in the inode cache. This is a generalized version of ilookup() for
974 * file systems where the inode number is not sufficient for unique
975 * identification of an inode.
977 * If the inode is in the cache, the inode lock is waited upon and the inode is
978 * returned with an incremented reference count.
980 * Otherwise NULL is returned.
982 * Note, @test is called with the inode_lock held, so can't sleep.
984 struct inode
*ilookup5(struct super_block
*sb
, unsigned long hashval
,
985 int (*test
)(struct inode
*, void *), void *data
)
987 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
989 return ifind(sb
, head
, test
, data
, 1);
991 EXPORT_SYMBOL(ilookup5
);
994 * ilookup - search for an inode in the inode cache
995 * @sb: super block of file system to search
996 * @ino: inode number to search for
998 * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
999 * This is for file systems where the inode number is sufficient for unique
1000 * identification of an inode.
1002 * If the inode is in the cache, the inode is returned with an incremented
1005 * Otherwise NULL is returned.
1007 struct inode
*ilookup(struct super_block
*sb
, unsigned long ino
)
1009 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1011 return ifind_fast(sb
, head
, ino
);
1013 EXPORT_SYMBOL(ilookup
);
1016 * iget5_locked - obtain an inode from a mounted file system
1017 * @sb: super block of file system
1018 * @hashval: hash value (usually inode number) to get
1019 * @test: callback used for comparisons between inodes
1020 * @set: callback used to initialize a new struct inode
1021 * @data: opaque data pointer to pass to @test and @set
1023 * iget5_locked() uses ifind() to search for the inode specified by @hashval
1024 * and @data in the inode cache and if present it is returned with an increased
1025 * reference count. This is a generalized version of iget_locked() for file
1026 * systems where the inode number is not sufficient for unique identification
1029 * If the inode is not in cache, get_new_inode() is called to allocate a new
1030 * inode and this is returned locked, hashed, and with the I_NEW flag set. The
1031 * file system gets to fill it in before unlocking it via unlock_new_inode().
1033 * Note both @test and @set are called with the inode_lock held, so can't sleep.
1035 struct inode
*iget5_locked(struct super_block
*sb
, unsigned long hashval
,
1036 int (*test
)(struct inode
*, void *),
1037 int (*set
)(struct inode
*, void *), void *data
)
1039 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1040 struct inode
*inode
;
1042 inode
= ifind(sb
, head
, test
, data
, 1);
1046 * get_new_inode() will do the right thing, re-trying the search
1047 * in case it had to block at any point.
1049 return get_new_inode(sb
, head
, test
, set
, data
);
1051 EXPORT_SYMBOL(iget5_locked
);
1054 * iget_locked - obtain an inode from a mounted file system
1055 * @sb: super block of file system
1056 * @ino: inode number to get
1058 * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
1059 * the inode cache and if present it is returned with an increased reference
1060 * count. This is for file systems where the inode number is sufficient for
1061 * unique identification of an inode.
1063 * If the inode is not in cache, get_new_inode_fast() is called to allocate a
1064 * new inode and this is returned locked, hashed, and with the I_NEW flag set.
1065 * The file system gets to fill it in before unlocking it via
1066 * unlock_new_inode().
1068 struct inode
*iget_locked(struct super_block
*sb
, unsigned long ino
)
1070 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1071 struct inode
*inode
;
1073 inode
= ifind_fast(sb
, head
, ino
);
1077 * get_new_inode_fast() will do the right thing, re-trying the search
1078 * in case it had to block at any point.
1080 return get_new_inode_fast(sb
, head
, ino
);
1082 EXPORT_SYMBOL(iget_locked
);
1084 int insert_inode_locked(struct inode
*inode
)
1086 struct super_block
*sb
= inode
->i_sb
;
1087 ino_t ino
= inode
->i_ino
;
1088 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, ino
);
1090 inode
->i_state
|= I_NEW
;
1092 struct hlist_node
*node
;
1093 struct inode
*old
= NULL
;
1094 spin_lock(&inode_lock
);
1095 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1096 if (old
->i_ino
!= ino
)
1098 if (old
->i_sb
!= sb
)
1100 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1104 if (likely(!node
)) {
1105 hlist_add_head(&inode
->i_hash
, head
);
1106 spin_unlock(&inode_lock
);
1110 spin_unlock(&inode_lock
);
1112 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1119 EXPORT_SYMBOL(insert_inode_locked
);
1121 int insert_inode_locked4(struct inode
*inode
, unsigned long hashval
,
1122 int (*test
)(struct inode
*, void *), void *data
)
1124 struct super_block
*sb
= inode
->i_sb
;
1125 struct hlist_head
*head
= inode_hashtable
+ hash(sb
, hashval
);
1127 inode
->i_state
|= I_NEW
;
1130 struct hlist_node
*node
;
1131 struct inode
*old
= NULL
;
1133 spin_lock(&inode_lock
);
1134 hlist_for_each_entry(old
, node
, head
, i_hash
) {
1135 if (old
->i_sb
!= sb
)
1137 if (!test(old
, data
))
1139 if (old
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
))
1143 if (likely(!node
)) {
1144 hlist_add_head(&inode
->i_hash
, head
);
1145 spin_unlock(&inode_lock
);
1149 spin_unlock(&inode_lock
);
1151 if (unlikely(!hlist_unhashed(&old
->i_hash
))) {
1158 EXPORT_SYMBOL(insert_inode_locked4
);
1161 * __insert_inode_hash - hash an inode
1162 * @inode: unhashed inode
1163 * @hashval: unsigned long value used to locate this object in the
1166 * Add an inode to the inode hash for this superblock.
1168 void __insert_inode_hash(struct inode
*inode
, unsigned long hashval
)
1170 struct hlist_head
*head
= inode_hashtable
+ hash(inode
->i_sb
, hashval
);
1171 spin_lock(&inode_lock
);
1172 hlist_add_head(&inode
->i_hash
, head
);
1173 spin_unlock(&inode_lock
);
1175 EXPORT_SYMBOL(__insert_inode_hash
);
1178 * remove_inode_hash - remove an inode from the hash
1179 * @inode: inode to unhash
1181 * Remove an inode from the superblock.
1183 void remove_inode_hash(struct inode
*inode
)
1185 spin_lock(&inode_lock
);
1186 hlist_del_init(&inode
->i_hash
);
1187 spin_unlock(&inode_lock
);
1189 EXPORT_SYMBOL(remove_inode_hash
);
1192 * Tell the filesystem that this inode is no longer of any interest and should
1193 * be completely destroyed.
1195 * We leave the inode in the inode hash table until *after* the filesystem's
1196 * ->delete_inode completes. This ensures that an iget (such as nfsd might
1197 * instigate) will always find up-to-date information either in the hash or on
1200 * I_FREEING is set so that no-one will take a new reference to the inode while
1201 * it is being deleted.
1203 void generic_delete_inode(struct inode
*inode
)
1205 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1207 list_del_init(&inode
->i_list
);
1208 list_del_init(&inode
->i_sb_list
);
1209 WARN_ON(inode
->i_state
& I_NEW
);
1210 inode
->i_state
|= I_FREEING
;
1211 inodes_stat
.nr_inodes
--;
1212 spin_unlock(&inode_lock
);
1214 if (op
->delete_inode
) {
1215 void (*delete)(struct inode
*) = op
->delete_inode
;
1216 /* Filesystems implementing their own
1217 * s_op->delete_inode are required to call
1218 * truncate_inode_pages and clear_inode()
1222 truncate_inode_pages(&inode
->i_data
, 0);
1225 spin_lock(&inode_lock
);
1226 hlist_del_init(&inode
->i_hash
);
1227 spin_unlock(&inode_lock
);
1228 wake_up_inode(inode
);
1229 BUG_ON(inode
->i_state
!= I_CLEAR
);
1230 destroy_inode(inode
);
1232 EXPORT_SYMBOL(generic_delete_inode
);
1235 * generic_detach_inode - remove inode from inode lists
1236 * @inode: inode to remove
1238 * Remove inode from inode lists, write it if it's dirty. This is just an
1239 * internal VFS helper exported for hugetlbfs. Do not use!
1241 * Returns 1 if inode should be completely destroyed.
1243 int generic_detach_inode(struct inode
*inode
)
1245 struct super_block
*sb
= inode
->i_sb
;
1247 if (!hlist_unhashed(&inode
->i_hash
)) {
1248 if (!(inode
->i_state
& (I_DIRTY
|I_SYNC
)))
1249 list_move(&inode
->i_list
, &inode_unused
);
1250 inodes_stat
.nr_unused
++;
1251 if (sb
->s_flags
& MS_ACTIVE
) {
1252 spin_unlock(&inode_lock
);
1255 WARN_ON(inode
->i_state
& I_NEW
);
1256 inode
->i_state
|= I_WILL_FREE
;
1257 spin_unlock(&inode_lock
);
1258 write_inode_now(inode
, 1);
1259 spin_lock(&inode_lock
);
1260 WARN_ON(inode
->i_state
& I_NEW
);
1261 inode
->i_state
&= ~I_WILL_FREE
;
1262 inodes_stat
.nr_unused
--;
1263 hlist_del_init(&inode
->i_hash
);
1265 list_del_init(&inode
->i_list
);
1266 list_del_init(&inode
->i_sb_list
);
1267 WARN_ON(inode
->i_state
& I_NEW
);
1268 inode
->i_state
|= I_FREEING
;
1269 inodes_stat
.nr_inodes
--;
1270 spin_unlock(&inode_lock
);
1273 EXPORT_SYMBOL_GPL(generic_detach_inode
);
1275 static void generic_forget_inode(struct inode
*inode
)
1277 if (!generic_detach_inode(inode
))
1279 if (inode
->i_data
.nrpages
)
1280 truncate_inode_pages(&inode
->i_data
, 0);
1282 wake_up_inode(inode
);
1283 destroy_inode(inode
);
1287 * Normal UNIX filesystem behaviour: delete the
1288 * inode when the usage count drops to zero, and
1291 void generic_drop_inode(struct inode
*inode
)
1293 if (!inode
->i_nlink
)
1294 generic_delete_inode(inode
);
1296 generic_forget_inode(inode
);
1298 EXPORT_SYMBOL_GPL(generic_drop_inode
);
1301 * Called when we're dropping the last reference
1304 * Call the FS "drop()" function, defaulting to
1305 * the legacy UNIX filesystem behaviour..
1307 * NOTE! NOTE! NOTE! We're called with the inode lock
1308 * held, and the drop function is supposed to release
1311 static inline void iput_final(struct inode
*inode
)
1313 const struct super_operations
*op
= inode
->i_sb
->s_op
;
1314 void (*drop
)(struct inode
*) = generic_drop_inode
;
1316 if (op
&& op
->drop_inode
)
1317 drop
= op
->drop_inode
;
1322 * iput - put an inode
1323 * @inode: inode to put
1325 * Puts an inode, dropping its usage count. If the inode use count hits
1326 * zero, the inode is then freed and may also be destroyed.
1328 * Consequently, iput() can sleep.
1330 void iput(struct inode
*inode
)
1333 BUG_ON(inode
->i_state
== I_CLEAR
);
1335 if (atomic_dec_and_lock(&inode
->i_count
, &inode_lock
))
1339 EXPORT_SYMBOL(iput
);
1342 * bmap - find a block number in a file
1343 * @inode: inode of file
1344 * @block: block to find
1346 * Returns the block number on the device holding the inode that
1347 * is the disk block number for the block of the file requested.
1348 * That is, asked for block 4 of inode 1 the function will return the
1349 * disk block relative to the disk start that holds that block of the
1352 sector_t
bmap(struct inode
*inode
, sector_t block
)
1355 if (inode
->i_mapping
->a_ops
->bmap
)
1356 res
= inode
->i_mapping
->a_ops
->bmap(inode
->i_mapping
, block
);
1359 EXPORT_SYMBOL(bmap
);
1362 * With relative atime, only update atime if the previous atime is
1363 * earlier than either the ctime or mtime or if at least a day has
1364 * passed since the last atime update.
1366 static int relatime_need_update(struct vfsmount
*mnt
, struct inode
*inode
,
1367 struct timespec now
)
1370 if (!(mnt
->mnt_flags
& MNT_RELATIME
))
1373 * Is mtime younger than atime? If yes, update atime:
1375 if (timespec_compare(&inode
->i_mtime
, &inode
->i_atime
) >= 0)
1378 * Is ctime younger than atime? If yes, update atime:
1380 if (timespec_compare(&inode
->i_ctime
, &inode
->i_atime
) >= 0)
1384 * Is the previous atime value older than a day? If yes,
1387 if ((long)(now
.tv_sec
- inode
->i_atime
.tv_sec
) >= 24*60*60)
1390 * Good, we can skip the atime update:
1396 * touch_atime - update the access time
1397 * @mnt: mount the inode is accessed on
1398 * @dentry: dentry accessed
1400 * Update the accessed time on an inode and mark it for writeback.
1401 * This function automatically handles read only file systems and media,
1402 * as well as the "noatime" flag and inode specific "noatime" markers.
1404 void touch_atime(struct vfsmount
*mnt
, struct dentry
*dentry
)
1406 struct inode
*inode
= dentry
->d_inode
;
1407 struct timespec now
;
1409 if (inode
->i_flags
& S_NOATIME
)
1411 if (IS_NOATIME(inode
))
1413 if ((inode
->i_sb
->s_flags
& MS_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1416 if (mnt
->mnt_flags
& MNT_NOATIME
)
1418 if ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
))
1421 now
= current_fs_time(inode
->i_sb
);
1423 if (!relatime_need_update(mnt
, inode
, now
))
1426 if (timespec_equal(&inode
->i_atime
, &now
))
1429 if (mnt_want_write(mnt
))
1432 inode
->i_atime
= now
;
1433 mark_inode_dirty_sync(inode
);
1434 mnt_drop_write(mnt
);
1436 EXPORT_SYMBOL(touch_atime
);
1439 * file_update_time - update mtime and ctime time
1440 * @file: file accessed
1442 * Update the mtime and ctime members of an inode and mark the inode
1443 * for writeback. Note that this function is meant exclusively for
1444 * usage in the file write path of filesystems, and filesystems may
1445 * choose to explicitly ignore update via this function with the
1446 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1447 * timestamps are handled by the server.
1450 void file_update_time(struct file
*file
)
1452 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1453 struct timespec now
;
1454 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
1456 /* First try to exhaust all avenues to not sync */
1457 if (IS_NOCMTIME(inode
))
1460 now
= current_fs_time(inode
->i_sb
);
1461 if (!timespec_equal(&inode
->i_mtime
, &now
))
1464 if (!timespec_equal(&inode
->i_ctime
, &now
))
1467 if (IS_I_VERSION(inode
))
1468 sync_it
|= S_VERSION
;
1473 /* Finally allowed to write? Takes lock. */
1474 if (mnt_want_write_file(file
))
1477 /* Only change inode inside the lock region */
1478 if (sync_it
& S_VERSION
)
1479 inode_inc_iversion(inode
);
1480 if (sync_it
& S_CTIME
)
1481 inode
->i_ctime
= now
;
1482 if (sync_it
& S_MTIME
)
1483 inode
->i_mtime
= now
;
1484 mark_inode_dirty_sync(inode
);
1485 mnt_drop_write(file
->f_path
.mnt
);
1487 EXPORT_SYMBOL(file_update_time
);
1489 int inode_needs_sync(struct inode
*inode
)
1493 if (S_ISDIR(inode
->i_mode
) && IS_DIRSYNC(inode
))
1497 EXPORT_SYMBOL(inode_needs_sync
);
1499 int inode_wait(void *word
)
1504 EXPORT_SYMBOL(inode_wait
);
1507 * If we try to find an inode in the inode hash while it is being
1508 * deleted, we have to wait until the filesystem completes its
1509 * deletion before reporting that it isn't found. This function waits
1510 * until the deletion _might_ have completed. Callers are responsible
1511 * to recheck inode state.
1513 * It doesn't matter if I_NEW is not set initially, a call to
1514 * wake_up_inode() after removing from the hash list will DTRT.
1516 * This is called with inode_lock held.
1518 static void __wait_on_freeing_inode(struct inode
*inode
)
1520 wait_queue_head_t
*wq
;
1521 DEFINE_WAIT_BIT(wait
, &inode
->i_state
, __I_NEW
);
1522 wq
= bit_waitqueue(&inode
->i_state
, __I_NEW
);
1523 prepare_to_wait(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
1524 spin_unlock(&inode_lock
);
1526 finish_wait(wq
, &wait
.wait
);
1527 spin_lock(&inode_lock
);
1530 static __initdata
unsigned long ihash_entries
;
1531 static int __init
set_ihash_entries(char *str
)
1535 ihash_entries
= simple_strtoul(str
, &str
, 0);
1538 __setup("ihash_entries=", set_ihash_entries
);
1541 * Initialize the waitqueues and inode hash table.
1543 void __init
inode_init_early(void)
1547 /* If hashes are distributed across NUMA nodes, defer
1548 * hash allocation until vmalloc space is available.
1554 alloc_large_system_hash("Inode-cache",
1555 sizeof(struct hlist_head
),
1563 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1564 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1567 void __init
inode_init(void)
1571 /* inode slab cache */
1572 inode_cachep
= kmem_cache_create("inode_cache",
1573 sizeof(struct inode
),
1575 (SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|
1578 register_shrinker(&icache_shrinker
);
1580 /* Hash may have been set up in inode_init_early */
1585 alloc_large_system_hash("Inode-cache",
1586 sizeof(struct hlist_head
),
1594 for (loop
= 0; loop
< (1 << i_hash_shift
); loop
++)
1595 INIT_HLIST_HEAD(&inode_hashtable
[loop
]);
1598 void init_special_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
)
1600 inode
->i_mode
= mode
;
1601 if (S_ISCHR(mode
)) {
1602 inode
->i_fop
= &def_chr_fops
;
1603 inode
->i_rdev
= rdev
;
1604 } else if (S_ISBLK(mode
)) {
1605 inode
->i_fop
= &def_blk_fops
;
1606 inode
->i_rdev
= rdev
;
1607 } else if (S_ISFIFO(mode
))
1608 inode
->i_fop
= &def_fifo_fops
;
1609 else if (S_ISSOCK(mode
))
1610 inode
->i_fop
= &bad_sock_fops
;
1612 printk(KERN_DEBUG
"init_special_inode: bogus i_mode (%o) for"
1613 " inode %s:%lu\n", mode
, inode
->i_sb
->s_id
,
1616 EXPORT_SYMBOL(init_special_inode
);
1619 * Init uid,gid,mode for new inode according to posix standards
1621 * @dir: Directory inode
1622 * @mode: mode of the new inode
1624 void inode_init_owner(struct inode
*inode
, const struct inode
*dir
,
1627 inode
->i_uid
= current_fsuid();
1628 if (dir
&& dir
->i_mode
& S_ISGID
) {
1629 inode
->i_gid
= dir
->i_gid
;
1633 inode
->i_gid
= current_fsgid();
1634 inode
->i_mode
= mode
;
1636 EXPORT_SYMBOL(inode_init_owner
);